SWIR and thermal imaging sensors

Executive summary

Short-wave infrared (SWIR) and thermal imaging sensors convert non-visible optical radiation into electronic signals. SWIR usually means approximately 0.9–1.7 µm, where InGaAs lattice-matched to InP is the incumbent high-performance detector. Uncooled thermal imagers usually operate in the 8–14 µm long-wave infrared (LWIR) band using vanadium-oxide (VOx) or amorphous-silicon microbolometers.

This note is anchored in Tier 1 evidence: a granted microbolometer patent, US5286976A, and peer-reviewed reviews by Rogalski. The granted patent proves an early recorded claim, but it does not prove present-day commercial performance. Vendor marketing pages provide Tier 2 adoption and specification context. Company 10-K/20-F filings are used to test marketing claims against audited risk disclosure.

The principal finding is that SWIR/thermal sensor economics are dominated by specialized fab and packaging capability, U.S. export controls, and defense/industrial integrator demand, not by a single current patent. Strong public evidence supports the technical feasibility of uncooled VOx thermal sensors and InGaAs SWIR sensors, but margin, yield, and buyer concentration data are mostly disclosed only at segment level or not at all.

Method and source discipline

This note follows the requested source hierarchy:

• Tier 1 — used for technical/performance/IP claims: granted patents, peer-reviewed papers with DOI, physical-constant standards, regulator technical assessments.
• Tier 2 — used for adoption/deployment context: company technical docs, engineering blogs, SEC filings, named conference proceedings.
• Prohibited: SEO content farms, anonymous benchmark sites, AI-generated summaries, and secondary sources that merely quote an unretrievable patent or paper.

A granted patent is distinguished from an application. Only the granted patent is used for IP-moat claims, and even then only as evidence of recorded claims, not as proof of performance or commercial viability. Peer-reviewed journal papers are preferred over preprints; no arXiv-only source is load-bearing in this note.

Sensor physics and technology families

2.1 SWIR: 0.9–1.7 µm

SWIR sensors detect reflected or scene light in the short-wave infrared band. The band sits just beyond the silicon imaging cutoff at roughly 1.1 µm. For high-sensitivity SWIR detection, the incumbent detector is InGaAs lattice-matched to InP, which covers about 0.9–1.7 µm. Peer-reviewed reviews consistently identify InGaAs/InP as the standard high-sensitivity SWIR detector architecture for uncooled or thermoelectrically cooled operation.¹²

Alternatives include HgCdTe with composition tuning, germanium on silicon, quantum-well infrared photodetectors, and colloidal quantum dot photodetectors. Rogalski 2011 notes that detector choice is driven by operating temperature, wavelength, dark current, and array format requirements.² The Tier 2 Sony SWIR product page describes its “SenSWIR” image sensors as using InGaAs photodiodes; the page presents the product as enabling imaging across visible through SWIR.³

The key physical boundary is that silicon photodiodes have low quantum efficiency beyond about 1.1 µm because photon energy falls below the silicon bandgap. That makes SWIR a separate material stack. InGaAs is expensive because it must be grown on InP substrates and hybridized to readout integrated circuits.¹²

2.2 Thermal: 8–14 µm LWIR

Thermal imaging detects emitted blackbody radiation from objects near room temperature. The 8–14 µm atmospheric transmission window is the dominant uncooled thermal imaging band. Incumbent uncooled detectors are microbolometers: micromachined thermally isolated resistors whose resistance changes with absorbed infrared radiation. The most common temperature-sensitive material is vanadium-oxide (VOx); amorphous silicon is a competing material.²

The foundational granted patent US5286976A, assigned to Honeywell, describes a microbolometer structure with thermal isolation.⁴ The patent was granted in 1994 and is now long expired. Its value here is as evidence of the early recorded microbolometer architecture, not as a current moat.

A typical uncooled VOx camera core is sold by Teledyne FLIR as the Boson family. Teledyne FLIR’s Boson product page describes an uncooled LWIR camera core with “<50 mK” noise-equivalent temperature difference (NETD) sensitivity and compact form factor.⁵ That is a self-reported vendor specification, not an independent test.

Microbolometer performance arises from the temperature coefficient of resistance (TCR) of VOx, commonly cited in the literature at approximately 2%/K for VOx.² The microbolometer is thermally isolated by a support structure and read out with a CMOS readout integrated circuit. Unlike cooled photon detectors, microbolometers operate at room temperature; the penalty is lower specific detectivity and slower response time than cooled HgCdTe or InSb photon detectors.²

Granted patent evidence

The note uses US5286976A as the Tier 1 granted patent.⁴

• Grant number: US5286976A
• Grant date: 1994-02-15
• Assignee: Honeywell Inc.
• Nature: microbolometer focal-plane element
• URL: patents.google.com

This is a granted patent, not an application. It is a foundational microbolometer patent, but it is expired and cannot support a present-day moat claim on its own. The patent should be read as proof that Honeywell recorded a microbolometer structure in the early 1990s, not as proof that any current commercial device meets today’s performance of <50 mK NETD or scales to high-volume production.

No current, unexpired granted patent is treated as load-bearing in this note because no current grant was verified against a primary patent-office record. The absence of a cited current grant does not mean modern SWIR/thermal sensor companies lack patents; it means this note does not assert a current patent moat without primary verification.

Peer-reviewed literature cross-check

Two Rogalski reviews are used as Tier 1 literature:

1. Rogalski, A., “Infrared detectors: status and trends,” Reports on Progress in Physics 74 (2011) 066501.

DOI: 10.1088/0034-4885/74/6/066501

URL: iopscience.iop.org

1. Rogalski, A., “Infrared detectors: status and trends,” Progress in Quantum Electronics 27 (2003) 59–210.

DOI: 10.1016/S0079-6727(02)00024-0

URL: doi.org

These are peer-reviewed journal publications with DOIs. They cover the detector physics used here: InGaAs and HgCdTe photon detectors for SWIR, VOx microbolometers for uncooled LWIR, and the trade-offs between cooled photon detectors and uncooled thermal detectors.¹²

For a load-bearing technical claim, the note requires two independent lineages. The microbolometer claim is cross-verified by the Rogalski reviews and the granted Honeywell patent. The InGaAs SWIR band claim is cross-verified by Rogalski and the Sony product page. The vendor-specific NETD and pixel specifications are single-source unless independently retested.

Company filings vs. marketing claims

5.1 Sony SWIR marketing

Sony Semiconductor Solutions markets SWIR image sensors under the “SenSWIR” line. The product page claims SWIR imaging with visible-light extension, and describes product benefits such as high resolution and industrial inspection ability.³ The page is a Tier 2 vendor tech document; it does not provide independent test data.

Sony’s SEC 20-F filing includes risk-factor language about the semiconductor and components segment. Sony’s 20-F is accessible through the SEC EDGAR company search page.⁶ The filing-level risk disclosure describes demand cyclicality, technology transitions, customer acceptance, and supply-chain factors. It does not isolate SWIR sensor yield, cost, or margin. Therefore, marketing specifications must be treated as target specifications rather than audited financial or production commitments.

Discrepancy observed: Sony’s marketing does not quantify yield or warranty performance; Sony’s 20-F risk factors do not provide SWIR-specific disclosure. The mismatch is one of granularity, not direct contradiction. The technical claims remain Tier 2 until independently measured.

5.2 Teledyne FLIR thermal camera cores

Teledyne FLIR’s Boson product page describes an uncooled thermal camera core with a vendor-stated NETD of <50 mK.⁵ Teledyne Technologies’ 10-K discloses segment-level results for its Digital Imaging segment, which includes FLIR. The 10-K is accessible through SEC EDGAR.⁷ The 10-K risk-factor language includes manufacturing and supply constraints, export controls, and rapid technological change. It does not break out microbolometer yield or camera-core unit economics.

Discrepancy observed: The Boson specification is vendor-reported and not independently certified on the product page. The 10-K does not specifically warrant the <50 mK NETD figure. Investors should not treat the marketing spec as an audited performance guarantee.

Quantitative worked calculation and sensitivity

6.1 Photon energy at 1.55 µm

SWIR fiber-communication and LIDAR wavelengths are near 1.55 µm. Photon energy is calculated with NIST CODATA constants.⁸

Formula:

\[

E_{\text{ph}} = \frac{h c}{\lambda}

\]

Inputs:

• \(h = 6.62607015 \times 10^{-34}\ \text{J}\cdot\text{s}\)
• \(c = 299792458\ \text{m/s}\)
• \(\lambda = 1.55\ \mu\text{m} = 1.55 \times 10^{-6}\ \text{m}\)

Calculation:

\[

h c = 1.986445857 \times 10^{-25}\ \text{J}\cdot\text{m}

\]

\[

E_{\text{ph}} = \frac{1.986445857 \times 10^{-25}}{1.55 \times 10^{-6}} = 1.28158 \times 10^{-19}\ \text{J}

\]

Convert to electronvolts using \(1\ \text{eV} = 1.602176634 \times 10^{-19}\ \text{J}\):

\[

E_{\text{ph}} = 0.800\ \text{eV}

\]

For an optical power of \(1\ \mu\text{W} = 10^{-6}\ \text{W}\):

\[

\text{photon rate} = \frac{10^{-6}\ \text{W}}{1.28158 \times 10^{-19}\ \text{J}} = 7.80 \times 10^{12}\ \text{photons/s}

\]

Sensitivity: If the wavelength is changed by ±0.05 µm, photon energy changes by about ∓3.2%, and the photon count per fixed optical power changes by the inverse proportion. This is purely physical and grounded in NIST constants.⁸

6.2 Microbolometer resistance change from a 50 mK scene temperature difference

Using a literature-approximate VOx TCR of \(\alpha = 2\%/\text{K}\), cited from the Rogalski review discussion of VOx properties, the fractional resistance change for a 50 mK differential target is:

\[

\frac{\Delta R}{R} = \alpha \Delta T = 0.02\ \text{K}^{-1} \times 0.05\ \text{K} = 0.001

\]

That is a 0.1% resistance change. This small change illustrates the precision required in readout electronics and is consistent with the Tier 1 physics description of microbolometers.²

Sensitivity: If TCR is 1.5%/K instead of 2%/K, the resistance change is 0.075%; if TCR is 2.5%/K, it is 0.125%. The calculation is illustrative; exact TCR depends on the specific VOx deposition process and is not disclosed on the Teledyne FLIR Boson page.⁵

Parameter table

Unit economics / business model

The business model differs by position in the value chain.

Sensor and camera-core vendors monetize through chip/module sales, not subscriptions or per-image royalties. Examples:

• Sony Semiconductor Solutions sells SWIR image sensors to industrial and machine-vision customers.³
• Teledyne FLIR sells uncooled thermal camera cores such as Boson to integrators and OEMs.⁵
• onsemi sells industrial image sensors, including SWIR devices, to machine-vision camera makers.⁹

Teledyne reports segment-level financial results for its Digital Imaging segment, which includes FLIR, through its 10-K.⁷ The filing does not disclose SWIR-specific gross margin or microbolometer wafer yield. Any specific margin figure would require primary-source extraction from the filing; currently, segment-level margin disclosure exists but sensor-level margin data is not publicly isolated. Therefore, no specific margin number is asserted in this note without a label of “needs primary source validation.”

Defense primes and integrators capture value through program-level contracts, where the sensor is embedded in a larger system such as a targeting pod, missile warning system, or vehicle sight. Their filings show exposure through contract and supply-chain risk.⁷

The economic bottleneck is not necessarily the SWIR/thermal pixel design; it is compound-semiconductor fab capacity, hybridization between detector and ROIC, vacuum packaging, and export-control compliance. The Tier 1 literature confirms that fabrication complexity separates uncooled microbolometers from standard CMOS image sensors.²

Disconfirming Evidence

The SWIR/thermal sensor investment thesis would be weakened or invalidated by:

• Silicon-based SWIR alternatives reaching adequate detectivity with standard CMOS-like cost, because that would remove the InGaAs/InP fab barrier.
• Colloidal quantum dot or germanium SWIR arrays passing automotive reliability and hot-pixel stability tests, because they could commoditize SWIR pricing.
• Export-control or supply-chain restrictions on InP/InGaAs substrates or HgCdTe materials, because those would constrain unit delivery despite strong demand.
• Failure of uncooled microbolometer pixel pitch scaling below the incumbent 12–17 µm range, because it would cap resolution and reduce consumer/automotive adoption.
• Evidence that current SWIR/thermal patent portfolios are mostly applications rather than granted, enforceable claims, which would weaken IP-moat assumptions even though patents are never proof of performance.
• Company 10-K disclosures showing persistent write-downs or restructuring in digital imaging/infrared segments, which would indicate that marketing growth claims are not converting to audited profitability.